Intel i86W Manual page 181

PROGRAMMING EXAMPLES
II
MATRIX MULTIPLY, C = A
*
8, CACHED AND PIPELINED LOADS MIXED
II
Registers loaded by calling routine
A=r16 II pointer into A, stored in memory by rows
8=r17
II
pointer into 8, stored in memory by columns
C=r18 II pointer into C, stored inmemory by rows
L=r19
II
the number of rows in A
M=r20 II the number of columns in A and rows in 8
N=r21 II the number of columns in 8
II Registers used locally
Ap=r29
II
temporary pointer into A
RC=r28 II rowlcolumn counter decremented by bla for loop control
DEC=r27 II decrement or for rowlcolumn pointers
Ar=r26
II
counter of rows in A
8c=r2S
II
counter of columns in 8
8p=r24
II
temporary pointer into 8
SIZ=r23
II
number of bytes in row of A or column of 8
Al=f4; A2=fS; A3=f6; A4=f7; AS=f8; A6=f9; A7=f10;A8=fll
II
matrix A row values
81=f12;82=f13;83=f14;84=flS;8S=f16;86=f17;87=f18;88=fl9
II
matrx 8 column vals
Tl=f20;T2=f21;T3=f22
II
temporary results
mov 8,
8p
II
Pointer to 8
shl 2, M, SIZ
II
Number of bytes in M entries
adds -8, r0, DEC II
Set decrement or for bla
adds -8, M, RC II Initialize rowlcolumn counter
d.fiadd·dd f0, f0, f0 II
Initiate dual-instruction mode
adds -4,
C,
C
II
Start C index one entry low
d.fnop II First dual-mode pair
adds -1, L, Ar II
Make row counter zero relative
d.fnop
II
bla DEC, RC, start_row II
Initialize LCC
d.fnop
II
mov A,
Ap
II
Pointer to A
start_row:: II
Executed once per row of A
d.pfmul.ss f0,
f0, f0
II
pfld. d
o
(8p) ,
f0
II
Load 2 entries of 8 into load pipe
d.pfmul.ss f0,
f0, f0
II
pfld.d 8(8p)++, f0
II Load 2 entries of 8 into load pipe
d.pfmul.ss f0,
f0, f0
II
pfld. d 8(8p)++, f0
II Load 2 entries of 8 into load pipe
d.pfadd.ss f0,
f0, f0
II
fld.q
o
(Ap) ,
Al
II
Load entries of A
d.pfadd.ss
f0, f0, f0 II
pfld.d 8(8p)++, 81 II
Load 2 entries of 8
d.pfadd.ss f0, f0, f0 II
adds -1, N, 8c
II
Initialize column counter
d.fnop II
pfld.d 8(8p)++, 83 II
Load 2 entries of 8
inner_loop:: II
Process eight entries from row of A with eight from col of 8
d.m12apm.ss Al, 81, f0 II
fld.q
16(Ap)++, AS
II
Load
entries of A
d.m12apm.ss A2, 82, f0 II
pfld.d 8(8p)++, 85
II Load 2 entries of 8
d.m12apm.ss A3, 83, f0 II
pfld.d 8(8p)++, 87 II
Load 2 entries of 8
Ex~mple
9-14. Matrix Multiply, Cached and Pipelined Loads (Sheet 1 of 2)
9-19